Instructions to use jedisct1/MiMo-V2.5-coder-Q2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use jedisct1/MiMo-V2.5-coder-Q2 with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="jedisct1/MiMo-V2.5-coder-Q2",
	filename="MiMo-V2.5-coder-Q2-00001-of-00016.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use jedisct1/MiMo-V2.5-coder-Q2 with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf jedisct1/MiMo-V2.5-coder-Q2
# Run inference directly in the terminal:
llama-cli -hf jedisct1/MiMo-V2.5-coder-Q2

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf jedisct1/MiMo-V2.5-coder-Q2
# Run inference directly in the terminal:
llama-cli -hf jedisct1/MiMo-V2.5-coder-Q2

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf jedisct1/MiMo-V2.5-coder-Q2
# Run inference directly in the terminal:
./llama-cli -hf jedisct1/MiMo-V2.5-coder-Q2

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf jedisct1/MiMo-V2.5-coder-Q2
# Run inference directly in the terminal:
./build/bin/llama-cli -hf jedisct1/MiMo-V2.5-coder-Q2

Use Docker

docker model run hf.co/jedisct1/MiMo-V2.5-coder-Q2

LM Studio
Jan

vLLM

How to use jedisct1/MiMo-V2.5-coder-Q2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jedisct1/MiMo-V2.5-coder-Q2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jedisct1/MiMo-V2.5-coder-Q2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/jedisct1/MiMo-V2.5-coder-Q2

Ollama
How to use jedisct1/MiMo-V2.5-coder-Q2 with Ollama:
```
ollama run hf.co/jedisct1/MiMo-V2.5-coder-Q2
```

Unsloth Studio

How to use jedisct1/MiMo-V2.5-coder-Q2 with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for jedisct1/MiMo-V2.5-coder-Q2 to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for jedisct1/MiMo-V2.5-coder-Q2 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for jedisct1/MiMo-V2.5-coder-Q2 to start chatting

How to use jedisct1/MiMo-V2.5-coder-Q2 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf jedisct1/MiMo-V2.5-coder-Q2

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "jedisct1/MiMo-V2.5-coder-Q2"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use jedisct1/MiMo-V2.5-coder-Q2 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf jedisct1/MiMo-V2.5-coder-Q2

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default jedisct1/MiMo-V2.5-coder-Q2

Run Hermes

hermes

Docker Model Runner
How to use jedisct1/MiMo-V2.5-coder-Q2 with Docker Model Runner:
```
docker model run hf.co/jedisct1/MiMo-V2.5-coder-Q2
```

Lemonade

How to use jedisct1/MiMo-V2.5-coder-Q2 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull jedisct1/MiMo-V2.5-coder-Q2

Run and chat with the model

lemonade run user.MiMo-V2.5-coder-Q2-{{QUANT_TAG}}

List all available models

lemonade list

jedisct1 commited on 10 days ago

Commit

c55021c

verified ·

1 Parent(s): 1d3db99

Upload README.md with huggingface_hub

Browse files

Files changed (1) hide show

README.md +2 -2

README.md CHANGED Viewed

@@ -75,7 +75,7 @@ The first plain `Q2_K` family candidate was small enough, but it was not reliabl
 - embeddings and output tensors stay higher precision because they are important for token identity and exact syntax
 - attention tensors are protected because tool-call and code prompts are structure-heavy
 - the dense first FFN is protected because early-layer representation quality matters disproportionately after heavy quantization
-- MoE down-expert tensors use `Q3_K`; this was kept from the known-good imatrix recipe rather than isolated as the only required choice
 That is why this is still a Q2-class build, but not the smallest possible Q2 build.
@@ -107,7 +107,7 @@ The important point is not that these small harnesses prove universal coding abi
 ## Tool-calling validation
-Tool calling was exercised in realistic agent loops rather than only checking toy single-call examples. The harness used for this validation was [Swival](https://swival.dev). Nothing in the build is tied to it, and any OpenAI-compatible agent harness should work just as well.
 Validation included:

 - embeddings and output tensors stay higher precision because they are important for token identity and exact syntax
 - attention tensors are protected because tool-call and code prompts are structure-heavy
 - the dense first FFN is protected because early-layer representation quality matters disproportionately after heavy quantization
+- MoE down-expert tensors use `Q3_K`, which was a better quality/memory tradeoff than pushing all expert down-projections lower
 That is why this is still a Q2-class build, but not the smallest possible Q2 build.
 ## Tool-calling validation
+Tool calling was exercised in realistic agent loops rather than only checking toy single-call examples. The harness used for this validation was [Swival](https://swival.dev). Nothing in the build is tied to it, and any OpenAI-compatible agent harness is likely to work in much the same way, but Swival is the only one that has actually been put through its paces here.
 Validation included: